KiMONo

KiMONo infers multilevel association networks from integrated multi-omics datasets to elucidate associations among genomics, epigenetics, transcriptomics, proteomics, and clinical features that may underlie disease pathophysiology.

Key Features:

• Sparse-group-LASSO regression: Uses sparse-group-LASSO to handle high-dimensional data while enforcing sparsity and grouping effects.
• Prior knowledge integration: Incorporates prior biological knowledge such as protein-protein interactions from Biogrid to inform network inference.
• Multi-omics support: Integrates data across multiple molecular levels including mutation, epigenetics, transcriptomics, proteomics, and clinical information.
• Network representation: Constructs networks where nodes represent diverse omic features and edges indicate significant associations between features.
• Biomarker and signature detection: Identifies disease-specific omic features and molecular signatures from integrated datasets.
• Robustness to limited and noisy data: Demonstrates performance in low-sample-size datasets and in the presence of noisy measurements.
• Applied validations: Has been applied to the Pan-cancer collection integrating five levels and to a major depressive disorder cohort integrating four levels, detecting expression quantitative trait methylation sites and loci and showing advantages over state-of-the-art methods.

Scientific Applications:

• Pan-cancer integrative analysis: Detects cancer-specific omic features by integrating mutation, epigenetics, transcriptomics, proteomics, and clinical data in the Pan-cancer collection.
• Major depressive disorder cohort analysis: Identifies expression quantitative trait methylation sites and loci from a four-level dataset combining genetic, epigenetic, transcriptional, and clinical data.
• Biomarker discovery: Reveals candidate biomarkers and molecular signatures associated with disease phenotypes across multiple omic layers.
• Disease mechanism elucidation: Infers networks that help elucidate associations potentially underlying disease pathophysiology.
• Analysis under limited data conditions: Applied to studies with low sample sizes and noisy data to extract robust multi-omic associations.

Methodology:

Network inference using sparse-group-LASSO regression to enforce sparsity and grouping effects, incorporation of prior biological knowledge (e.g., Biogrid protein-protein interactions), and construction of networks with nodes as omic features and edges representing significant associations.